Air Traffic Analysis, Inc

Using WITI for Airport Arrival Performance Analysis

A report on work-in-progress

December 2010

2

WITI and WITI-FA (“Forecast Accuracy”)

WITI = “Weather weighted by traffic”

• En-route weather• E-WITI uses actual convective Wx data, e.g. NCWD

• E-WITI-FA uses convective forecast data, e.g. CCFP, LAMP, …

• Both use the same scheduled traffic on major flows

• Convective forecast data is converted to “quasi-NCWD” format

• Terminal weather• T-WITI uses actual surface Wx data (METARs)

• T-WITI-FA uses surface Wx forecast data (TAFs)

• Both use the same scheduled traffic at major airports

• TAF converted to quasi-METAR form, “rolling look-ahead” stream

3

VFR

IFR

VFR

Weather Event

Arriv

al R

ates

Time

Airport Capacity Rate

Actual Arrivals

region of possible avoidable costs

Framework for Quantifying Avoidable and Unavoidable Weather Impact

4

Arrival Rate Deficit: Illustration

0

5

10

15

20

25

30

35

40

45

6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 0 1 2 3 4 5

AAR

Hour, Z

LGA, Apr 3, 2009 Arrival Rates

Actual arrivals

Arr rate based on METAR

Arr rate based on 4hr TAF

Scheduled arrivals

Example: LGA, 04/03/2009

Forecast called for rain, low ceilings and strong winds from the southwest which would have forced LGA into a single-runway operation with low arrival rate.

Actual winds were much weaker. Ceilings lifted earlier than forecast.

Deficit between scheduled and actually-achieved arrival rates needs to be measured

Portion attributable to inaccurate weather forecast needs to be quantified

This portion then needs to be split into two pieces: see next slide

5

Accidents, outages, VIP flights, security, etc

0

5

10

15

20

25

30

35

40

45

6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 0 1 2 3 4 5

AAR

Hour, Z

LGA, Apr 3, 2009 Arrival Rates

Actual arrivals

Arr rate based on METAR

Arr rate based on 4hr TAF

Scheduled arrivals

Avoidable

UnavoidableReasonable risk mgmt

Impact caused by other airports

Inaccurate forecast

Overly conservative TMI

Actual AAR < TMI (in excess of risk mgmt)

Deficit

Bad Weather

Avoidable and Unavoidable Portions of Arrival Rate Deficit: Breakdown

Unavoidable portion of arrival rate deficit (gap between “scheduled” and “maximum-achievable-given-actual-Wx” arrival rates) needs to be subtracted from overall deficit.

A reasonable risk mitigation factor (% arrival rate?) should also be subtracted.

Impact caused by other airports, unrelated to this one, should be subtracted as well.

What’s left is the avoidable portion. It can be subdivided into 4 categories: (a) Deficit caused by inaccurate forecast; (b) Overly conservative TMI – related to (a) and possibly partly caused by it; (c) Actual AAR below TMI’s (even after discounting for risk factor), and (d) Transition.

En-route to Terminal to Final Transition

6

Adding GDP InformationGDP and non-GDP Periods

0

10

20

30

40

50

60

70

80

90

100

0

10

20

30

40

50

60

6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 0 1 2 3 4 5

Avoi

dabl

e D

elay

AAR

Hour, Z

EWR, Sep 30, 2010 Arrival Rates

Avoidable Delay

Actual arrivals

Arr rate based on Obs

Arr rate based on 4hr TAF

Scheduled arrivals

GDP Rate

Non-GDP

GDP

Non-GDPGDP

7

Non-GDP Arrival Rate Deficit Dissection

0

10

20

30

40

50

60

70

80

90

100

0

10

20

30

40

50

60

6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 0 1 2 3 4 5

Avoi

dabl

e D

elay

AAR

Hour, Z

EWR, Sep 30, 2010 Arrival Rates

Avoidable Delay

Actual arrivals

Arr rate based on Obs

Arr rate based on 4hr TAF

Scheduled arrivals

GDP Rate

Not counted (scheduled arrival rate is too low)

Avoidable delay = 0 (actual arr rate >= scheduled)

Over-forecast and actual arr rate < scheduled: • all the deficit goes toward avoidable

delay (“Non-GDP Inefficiencies”);

• a portion of the deficit is attributed to Wx forecast inaccuracy

-100

0

100

200

300

400

500

600

700

800

More Onus on GDP Setup Less Onus on GDP Setup

Avoi

dabl

e D

elay

, Hou

rs

Avoidable Delay Estimates Including GDP Efficiency and GDP Execution, EWR, 09/30/2010

Non-GDP Inefficiencies

Non-GDP Wx Fcst Inaccuracy

GDP Wx Forecast Inaccuracy

GDP Setup Inefficiencies

GDP Execution Inefficiencies

Effect of Other Airports

GDP assessment includes a 7% risk management AAR credit(reduces avoidable delays)

Other airports' departure delaysare converted to arrival delaysfor affected destinations butonly for those hours when theiravoidable delays were > 0

Non-GDP I neff

Non-GDP Wx FcstInacc

Non-GDP

8

GDP Arrival Rate Dissection

0

10

20

30

40

50

60

70

80

90

100

0

10

20

30

40

50

60

6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 0 1 2 3 4 5

Avoi

dabl

e D

elay

AAR

Hour, Z

EWR, Sep 30, 2010 Arrival Rates

Avoidable Delay

Actual arrivals

Arr rate based on Obs

Arr rate based on 4hr TAF

Scheduled arrivals

GDP Rate

No impact (GDP > scheduled rate)

GDP Setup Inefficiency (but avoidable delay due to GDP Wx forecast inaccuracy = 0)GDP Execution

Inefficiencies

-100

0

100

200

300

400

500

600

700

800

More Onus on GDP Setup Less Onus on GDP Setup

Avoi

dabl

e D

elay

, Hou

rs

Avoidable Delay Estimates Including GDP Efficiency and GDP Execution, EWR, 09/30/2010

Non-GDP Inefficiencies

Non-GDP Wx Fcst Inaccuracy

GDP Wx Forecast Inaccuracy

GDP Setup Inefficiencies

GDP Execution Inefficiencies

Effect of Other Airports

GDP assessment includes a 7% risk management AAR credit(reduces avoidable delays)

Other airports' departure delaysare converted to arrival delaysfor affected destinations butonly for those hours when theiravoidable delays were > 0

GDP Execution Ineff

GDP

9

Avoidable Delay/Cost analysis work is in progress and will continue in FY11

Another WITI application: airport delay prediction

(Both programs are funded by the FAA ATO-P Aviation Weather Group and led by AvMet Applications, Inc)

10

Training WITI Model on Airport DelayExample: ATL, 2007 (WITI based on actual Wx)

0

20000

40000

60000

80000

100000

120000

140000

160000

180000

200000

ATL ASPM Delay (Daily Total), minutes, and ATL Airport WITI, 2007

WITI

Delay

Once trained on historical data, the Airport WITI model can be used for delay prediction using forecast Wx and scheduled traffic for the day

11

Hourly Delay Forecast, 06/08/2008 Forecast: LAMP (convective); TAF (surface)

0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 0 1 2 3 4 5

Delay, min

Hour, Z

EWR Actual and Forecast Delay by hour, 06/08/2008

Actual ASPM Delay, min

4hr WITI-FA forecast "Delay", min

12

0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 0 1 2 3 4 5

Delay, min

Hour, Z

EWR Actual and Forecast Delay by hour, 01/28/2009

Actual ASPM Delay, min

4hr WITI-FA forecast "Delay", min

Hourly Delay Forecast, 01/28/2009 Forecast: TAF (surface)

13

Back-up Slides

14

WITI: Measuring Weather / Traffic Impact “The Hand the NAS Is Dealt Every Day”

Traffic Demand

National Airspace System (NAS)

ATM, Airline Response Strategies

Operational Outcomes

Local Airport Weather

En-route Convective Weather

The Weather Index (WITI) expresses severity of weather impact on the NAS, weighted by air transportation service demands

Capacity, Safety constraints

15

Queuing Delay Modeled by Wx Index Software, PHL, Dec 14, 2006

0

30

60

90

120

150

0600Z Dec 14 - 0600Z Dec 15, Hourly Wx Observations

Ho

urly

Dem

and

, Est

imat

ed C

apac

ity,

an

d D

elay

Dep+Arr demand

Dep+Arr capacity

Queuing Delay, Hrs

Actual queuing delays at PHL on 12/14/06 were significant (223 flights delayed > 15 min) and there were approx. 100 cancellations (so the resulting queuing delay was less than it could have been if there were no cancellations)

Optimum capacity in good Wx

WITI is a weighted sum of three components:

WITI CompositionWeather Weighted by Traffic, Quantified

– En-route Component: hourly frequency on major flows X amount of convective Wx that these flows cross

– Terminal Component

15

Used by the FAA and NWS on a regular basis:• Measure system performance in an objective manner – weekly reports • Compare different seasons’ Wx/traffic impact with outcomes (e.g. delays)

– Linear part: capacity degradation due to terminal weather impact, proportional to number of ops

– Non-linear (Queuing Delay) part reflecting excess traffic demand vs. capacity

16

Method: Use Airport Arrival RatesCompute Arrival Rate Deficit

We compare:

· Scheduled arrival rates from ASPM database

· Actual arrival rates, also from ASPM

· Model-generated arrival capacity based on METARs (i.e., actual weather data)

· Model-generated rates capacity based on TAFs (i.e., forecast weather data)

· Computed using a parametric model of airport capacity under different Wx conditions

· Use FAA’s airport capacity benchmarks and historical data on actual airport throughput

Any arrival rate deficit (“possible minus actual”) may be an indication of avoidable delays / cancellations

17

Re-Tooling WITI as an Airport Model

Standard WITI is a NAS Wx Impact assessment tool

A weighted sum of 3 components

– Weights computed to provide best correlation between WITI and Delay for OEP34 airports combined

WITI can be re-tooled as an airport model / delay predictor

– Use an airport specific, much more detailed WITI metric and “train” it on that airport’s delay-vs-Wx-and-traffic-demand data

– 12 components instead of 3 (“ATL Wind WITI, EWR Snow WITI, ORD Convective WITI”, etc)

Calibrate WITI straight to minutes-of-delay for direct comparison with actual ASPM delays